Data-driven wind farm power forecasting with Numerical weather predictions and SCADA data
I tiakina i:
| I whakaputaina i: | Journal of Physics: Conference Series vol. 3143, no. 1 (Dec 2025), p. 012012 |
|---|---|
| Kaituhi matua: | |
| Ētahi atu kaituhi: | , , , , , , , , |
| I whakaputaina: |
IOP Publishing
|
| Ngā marau: | |
| Urunga tuihono: | Citation/Abstract Full Text - PDF |
| Ngā Tūtohu: |
Kāore He Tūtohu, Me noho koe te mea tuatahi ki te tūtohu i tēnei pūkete!
|
| Whakarāpopotonga: | Wind energy generation is becoming increasingly significant in the clean energy transition. On the one hand, enhancing the overall efficiency of wind farms presentsan obvious challenge to the broader adoption of this technology in the power productionlandscape. On the other hand, due to the nondeterministic nature of the primary energysource, improving the reliability of wind power plants is essential to ensure their effective and sustainable integration into the energy network. In this scenario, access to accurate power forecasts allows more effective planning of electrical load management, greater profits in the day-ahead energy market, and could improve the efficiency of operation and maintenance due to better planning of unit commitment and scheduling by system operators. However, the non-stationarity, randomness, and intermittency of the wind source make wind power forecasting particularly challenging. To address this challenge, this work presents a data-driven power forecasting framework to handle turbine-level time series collected from SCADA systems, capturing the operational state of the turbines and numerical weather prediction data, available on a grid of points centered on the wind farm. The latter, in addition to being inherently affected by uncertainty, are also available at locations that do not coincide with the actual positions of the turbines. As a result, the framework aims to build a spatial transfer function capable of transporting the forecast information to the exact points where the turbines are installed, while also incorporating SCADA-based operational data, leveraging real-time turbine status indicators. In this work, different model architectures are explored and applied to different subsets of input variables to generate multi-horizon forecasts with a maximum lead time of 24 hours. Furthermore, the approach is tested both by considering multiple turbinesand training a separate model for each, as well as by training a single farm-level model. |
|---|---|
| ISSN: | 1742-6588 1742-6596 |
| DOI: | 10.1088/1742-6596/3143/1/012012 |
| Puna: | Advanced Technologies & Aerospace Database |